On Thu, 25 Sep 2003 14:56:52 GMT, "seferiad" <[EMAIL PROTECTED]>
wrote:

> I want to compare the variance of two "different" data sets for hypothesis
> testing.  If I assume the distributions of both data sets were normal, then
> I simply apply the F-test.  But I do not want to assume that the
> distributions were derived from normal populations.

If you want to compare the  *variances*, the simplest F-test
uses the ratio of the variances.  That is sensitive, as you 
suggest (if that is what you are suggesting) to non-normality.
The usual alternative is to compute the absolute deviations 
from the median (or from the mean, for convenience)
and do a simple ANOVA on those numbers.  Levene test.

I wonder at the comment, about applying the F-test after
assuming the normality;  tests to compare sample variances  
are not of ordinary interest, except for ruling out problems 
with assumptions.   I would never use the ratio-test in place
of the Levene test, except as an informal guide while in
the midst of exploratory analyses.
 

> 
> I've searched and searched, but can't seem to find a clear explanation of
> what is the appropriate replacement test (for the F test) for two non-normal
> distributions.  Can anyone help?  Why is there so much discussion about
> U-MannWhitney and other non-parametric tests that seem to focus more on
> T-test replacements.  At least that is what it seems like from my limited
> perspective.

The easiest tests, on two or more groups, are 
equivalent to performing a rank-transformation on the
combined data; followed by ANOVA (or t-test) on the
result:  for small samples, the 'exact'  results could be 
obtained by tabulation, if that were desired.  - See the
1999   edition of Conover's  Practical Nonparametric
Statistics.

Textbooks and articles focus on the two-group 
situation (t-test)  because that is most common, and that
is the easiest to describe and illustrate.

If I am missing the point...  please clarify.
-- 
Rich Ulrich, [EMAIL PROTECTED]
http://www.pitt.edu/~wpilib/index.html
"Taxes are the price we pay for civilization." 
.
.
=================================================================
Instructions for joining and leaving this list, remarks about the
problem of INAPPROPRIATE MESSAGES, and archives are available at:
.                  http://jse.stat.ncsu.edu/                    .
=================================================================

Reply via email to